We present an extensive electromagnetic induction imaging (EMI) survey across a 66 ha catchment to quantify spatial variations in soil hydraulic properties controlling surface-groundwater interactions. The proposed methodology involves three steps: (1) inversion of EMI data using a deep learning (DL) network that permits rapid prediction of 1D electrical conductivity (EC) depth models, (2) correlation of predicted electrical conductivity (EC) with soil textural information and (3) prediction of saturated hydraulic conductivity (K_s) from the predicted soil textural information using a recalibrated pedotransfer function (PTF) developed for the catchment. The performance of the DL inversion is evaluated through comparison with the classical deterministic inversion approach, with both methods yielding similar EC sections without significant discrepancies. The experimental petrophysical relationships between EC and soil textural information at 300 sampling locations revealed reasonably well-defined correlations with silt (R2=0.4) and clay (R2=0.33), but no correlation with sand and the predicted soil profiles capture measured trends, with minor discrepancies of 2-5%. The recalibrated PTF demonstrates effective performance (R2=0.52) in estimating K_s and was evaluated using split-sample validation. The predicted K_s maps reveal substantial variability in shallow soil hydraulic properties, likely influenced by agricultural practices, with a shift towards lower K_s values at greater depths. The proposed approach, combining DL inversion of EMI data, site-specific petrophysical relationships, and a field-scale PTF, provides a framework for predicting hydraulic properties from EMI data in catchments with heavy soils, enhancing understanding of runoff generation mechanisms and enabling improved land management strategies.